Proximity Forest 2.0: A new effective and scalable similarity-based classifier for time series